In this chapter, we will review the application of frustration theory to the explanation of another case of reward–nonreward intermittency – discrimination learning. We will consider the role of frustrative factors of arousal and suppression in the formation of discriminations, and the mediating action of anticipatory frustration and counterconditioning in the retardation and facilitation of discrimination learning. The discrimination learning we will consider involves separate experiences with or exposures to the discriminanda, which first the case of dispositional learning. We should therefore briefly review the history of the involvement of frustration in such “go/no-go” discriminations.

Frustrative factors in discrimination learning

The Hullian analysis of instrumental discrimination learning (Hull, 1950, 1952), which has been and continues to be influential, did not include an active, frustrative role for nonreinforcement. In broad outline, while Hull did maintain that the primary process involved was differential reinforcement, he held that its major function was (a) to neutralize the effect of the background or context stimuli that occur in the presence of both discriminanda, (b) to increase the power of the positive stimulus to evoke the (approach) response, and (c) to decrease the power of the negative stimulus to evoke that response. However, in Hull's theorizing the negative stimulus did not elicit avoidance of nonreward; rather, responses elicited by the S – permitted accumulation of an inhibitory state, reactive inhibition (IR), which, not offset by the growth of excitatory strength and particularly under massed-trial conditions, led to the extinction of responding to S –.

In D. O. Hebb's last book, Essays on Mind (1980), there is the following passage:

This statement, in Hebb's colorful prose, is an example, in phylogenetic terms, of a kind of thinking that, in its ontogenetic counterpart as well as in levels of functioning in the adult mammal, is seen increasingly in our field (e.g., Amsel & Stanton, 1980; Bitterman, 1960, 1975; Livesey, 1986; Schneirla, 1959; Wickelgren, 1979). My own point (Amsel, 1972b) was (and is) that there is a level of classical conditioning that is purely dispositional, involves implicit memory, and is less dependent on mediation than what is usually called Pavlovian conditioning, and that both levels involve a lesser degree of mediation than instrumental learning.

Introduction: Modals and language–thought issues

The ability to communicate through language is a benchmark of human competence. Speakers use language to communicate their knowledge and beliefs and to share their community's rules of social order and cultural norms. Modal expressions, because they encode the stances speakers take on propositions, are essential to these two language functions. Consider, for example, a proposition like John come. With the addition of a modal, such as can, may, must, or will, speakers can make a variety of statements about that proposition, from expressing their degree of certainty about the likelihood of the event it encodes to giving permission for the event to occur. If children are to become competent members of their language community, it is crucial that they master their language's modal system.

Because modal expressions encode notions of necessity, possibility, obligation, and permission, they are an important and revealing aspect of language to study with regard to language–thought issues. The use of modals in the epistemic sense (having to do with beliefs, attitudes, and knowledge states) seems to entail the ability to assess one's own knowledge state, to evaluate evidence, and to communicate those assessments to another whose state may be different from one's own. Such abilities, central though they are to a host of cognitive activities, have not regularly been accorded to children younger than 4. Yet children younger than 4 certainly produce modals in their spontaneous speech. Possibly children use these words solely for purposes of social regulation and description, that is, with deontic and dynamic meanings rather than epistemic ones (Palmer, 1979).

Introduction

Each of the contributions to this volume began from a simple insight: If we are to understand fully either language or thought, we must understand how they develop and interact in children's minds. Philosophers have long speculated about the origins of our distinctly human abilities to speak and reason complexly. Only relatively recently, however, has the issue been the subject of serious scientific study. We are now in the midst of an explosion of such research. In the past twenty years we have witnessed many exciting advances in the study of language and cognitive development. Now more than ever before, the utterances of small children are being gathered, scrutinized, and incorporated into broader theories of human cognition. The purpose of the present volume is to report and reflect on these advances. The product is a set of detailed and compelling models of language–thought relations (specifically, meaning– concept relations) as reflected in the developing mind of the child.

There are two quite different motivations for focusing on development as a window onto language–thought relations. One motivation is that children's behaviors provide clues to constraints on language and on thought. We can determine how languages and conceptual systems are constrained by examining the forms and meanings that children construct, and which errors they fail to make. All natural languages must be learnable by children; thus, children provide clues to what is universal, not only in syntax but in semantics as well.

A very different motivation for taking a developmental approach is that it can provide valuable clues to the complexities of the system eventually acquired.

The goal of this book is to explore the multiple relations between language and cognition in the course of first language acquisition. Children learn an enormous amount during the few years between birth and the time they reach school age. Among their accomplishments are the organization of their experiences into categories and the acquisition of a native language. How do they accomplish these critical tasks? How do their conceptual systems influence the structure of the languages they speak? How do linguistic patterns influence how they (and we) view reality? These questions – and the explorations of their answers in this volume – should be of interest to professionals and advanced students in the fields of language development, cognitive development, and cognition more generally.

We are grateful to many people for making this volume possible. The authors of the chapters deserve special thanks and congratulations for their outstanding contributions. Donna Pineau at the University of Michigan spent many hours typing portions of the book and ferrying manuscripts to the post office. Helen Wheeler, Julia Hough, Katharita Lamoza, Mary Racine, and the staff at Cambridge University Press were unfailingly helpful and patient in creating a book out of what had begun as just a thought. We thank Gail Gottfried and Erin Hartman for their careful help with proofreading and preparing the indexes. We also acknowledge the generous assistance of NICHD Grant HD-23378, a Spencer Fellowship from the National Academy of Education, and an award from the W. K. Kellogg Foundation through the Presidential Intiatives Fund of the University of Michigan for supporting S. Gelman during the preparation of the book.

Similarity has been cast both as hero and as villain in theories of cognitive processing, and the same is true for cognitive development. On the positive side, Rosch and her colleagues have suggested that similarity is an initial organizing principle in the development of categorization (e.g., Rosch, Mervis, Gray, Johnson, & Boyes-Braem, 1976), and Carey (1985) implicates a similarity mechanism in children's learning of the biological domain. It has also been suggested that similarity may play a role in word acquisition (Anglin, 1970; Bowerman, 1973, 1976; E. V. Clark, 1973; Davidson & Gelman, 1990; Gentner, 1982c). Others have taken a more pessimistic view, according to which similarity is either a misleading or at best an inferior strategy used as a last resort. Keil (1989), for example, posits that children begin with theories of the world and that similarity functions merely as a fallback strategy to be resorted to when theory fails.

A related issue is the course of development of similarity. Many researchers have suggested that children's use of similarity changes from an early and naiïve form to a later, more enlightened form.

This chapter focuses on the development of young children's communicative abilities as they manifest themselves through talk, that is, the use of language to receive and send messages. From a communicative perspective, the relationship between language and thought can be conceptualized of as one of medium (language) to content (thought). The preschool years are a period of rapid development in the domain of communication as young children establish and expand their knowledge base and language skills, as well as the social skills that are needed for successful interpersonal interactions. Some excellent reviews of the literature on the development of communication processes are available (Garvey, 1984; Mueller & Cooper, 1986; Shatz, 1983). Here we limit ourselves to a consideration of the literature that bears directly on the question guiding our present research: How do 2- and 3-year-old children learn to communicate verbally with one another?

An extensive literature describes early parent–child interactions and argues that such interactions are crucial for early language development (e.g., Bruner, 1983; Snow, 1977, 1979). However, there is very little in the literature that addresses the question of how the young child moves out of the intimate parent–infant exchanges and learns to talk with peers. The literature on parent–infant conversational exchanges documents the parent's expertise as a conversational partner. How do young children, accustomed to conversing with their parents, move into situations in which they must converse with others, particularly with others who, like themselves, are conversational novices? Even conversationally skilled adults find 2-year-olds (particularly those who are not their own children) to be extremely difficult conversational partners.

Slobin (1979) notes that human culture, social behavior, and thinking could not exist as we know them in the absence of language. Although we are inclined to agree with both Slobin's and Aldous Huxley's assessments of the importance of language in our lives, it is not entirely clear how best to define that importance. This dilemma has bedeviled thoughtful persons since at least the beginnings of philosophical inquiry. In this chapter we begin with a brief review of the language–thought controversy and an analysis of the roots of the problem, and then describe some research that addresses the issue.

A taxonomy of theories

Huxley and Slobin represent a traditional and persistent hypothesis in psychology, that language and thought are equivalent and that language is primary. In the behaviorist tradition, thought was considered to be subvocal speech. In contemporary psychology, knowledge is characterized as semantic understanding. A concept is a dictionary definition or an encyclopedia entry. Planning, problem solving, and reasoning, all examples of thought, require that knowledge used to access them be coded and stored. Although there is much debate about the form in which coding takes place, words are among the most obvious candidates. Contemporary cognitive psychology has proposed other alternatives that are also linguistic, such as propositions (Frederiksen, 1975; Kintsch, 1974), prototypes, and features. Propositional models interrelate concepts in an abstract network. The links between concepts in semantic networks are largely sentential. Feature models break down concepts into semantic components, such as human and female.

Current views of conceptual development assume that children actively construct mental categories. In this sense, the field owes much to the insight of Piaget. Although many researchers in the field of cognitive development have rejected Piaget's claims about qualitative stage changes in development, we almost take for granted his insight about children's active role in organizing and interpreting their experiences. Within the field, there is another growing trend toward considering the construction of knowledge not just as a pursuit of the individual, but as a social process. Much of this work has been inspired by Vygotsky's ideas. Piaget's and Vygotsky's approaches may seem at first to be at odds with one another, but in fact they are quite complementary. By combining these two approaches we may better understand the interaction between (a) children's expectations about category structure and word meaning and (b) the way that parents use language to structure the world for young children. Considerable attention is now focused on a third approach, which emphasizes not only the separate roles of parents and children, but also the convergence process by which parents and children construct shared meanings for words and concepts.

The notion that language is situated within social interaction and that conversational partners converge on the meanings of words within conversation has its roots in Vygotsky's (1962, 1978) work and has been elaborated recently in Rogoff (1990), Bruner (1984, 1987), H. Clark (1985), and others (see Bruner & Haste, 1987). Vygotsky (1962, 1978) described cognitive development as a process of internalizing patterns of social interaction (see Wertsch & Stone, 1985).

As the chapters in this book testify, the relation between language and thought in development has been conceptualized in many different ways. As the question of how children acquire language has come into focus in the past 25 years, theorists have argued either that language depends on cognition or the opposite – that cognition depends on language. This argument has taken many forms (for some of the forms and their ramifications see Cromer, 1988; Johnston, 1985). The classical views are reviewed elsewhere in this volume. In order to put my own conception of this relation into perspective it is revealing to consider briefly the many variations that are implicit in current psychological and linguistic models.

First, there is the strong cognitive determinism position attributed to Piaget – that acquiring and using language depends on the prior acquisition of supportive cognitive structures. Despite many efforts to find empirical support for this general proposition, there is little hard evidence that specific cognitive achievements are prerequisites of the acquisition of specific linguistic developments. Second, in the linguistic determinism position (known as the Sapir–Whorf hypothesis), thought, in its final forms, is held to be a function of the particular system of language that an individual has learned to use. Although both weak and strong forms of this position have been generally discredited in the recent past, a revival of sorts is under way (see Lakoff, 1987; Lucy, 1985; Silverstein, 1985). Third, the position that cognition is language – that the two are functionally equivalent – was implied in Watsonian behaviorism, which held thought to be covert verbalization.

What is striking about human categories is their diversity. They range from the simplest classification of a face or color to the most carefully constructed taxonomic grouping. Considering this diversity, many are struck by the apparent gap between the simple, intuitive categories formed by children and the complex, theory-laden categories of educated adults (e.g., Inhelder & Piaget, 1964; Quine, 1969; Vygotsky, 1962).

In this chapter we first argue that despite a number of salient differences between children's categories and those of adults, there are important parallels between the two: Both are informed by an ability to overlook salient appearances, an attention to nonobvious properties, and the potential to draw many new inferences about the unknown. Both the initial groupings of the prescientific child and the most thoughtful, theory-laden classifications of the adult extend knowledge in important ways. To use Quine's terminology, both children and adults form “theoretical kinds.” Second, we address the role of language in the formation of theoretical kinds. Although the structure of everyday object categories (e.g., dog, hammer, oak tree, and computer) is traditionally thought to result from the structure of the world and/or the nature of human perception and cognition, we will present evidence that language is also critical and that how objects are named helps determine the structure of the categories they fall into.

The chapter has three sections. In the first, we set forth our assumptions about the nature of categories for adults. We review recent analyses suggesting that categories are enriched and informed by intuitive theories, summarizing arguments from psychology, philosophy, and linguistics to converge on the point that human categories extend far beyond observable similarities.

Introduction

A new emphasis has emerged in the literature on concepts and concept acquisition. There is now a strong focus on the role of theories and other explanatory belief systems in structuring concepts and conceptual change. A series of recent studies on adult concepts (e.g., Medin & Wattenmaker, 1987; Murphy and Medin, 1985; Rips, 1989) and a rediscovery of older work, such as Asch (1946), Luchins (1957), and Chapman and Chapman (1969), as well as developmental analyses (e.g., Carey, 1985; Keil, 1986, 1989; Wellman & Gelman, 1988), reveal the vital importance of theorylike beliefs in understanding concept structure and development. At roughly the same time, and for subtly related reasons, there has been a resurgence of interest in domain specificity of knowledge in cognitive development, with the consequence that researchers are now wondering whether the development of knowledge in different theoretical domains might influence concepts in ways that are unique to each domain. This chapter assesses how these two trends link up to help us understand the acquisition of word meaning.

Although the mapping between concept structure and semantic structure is certainly not one to one, as has been clearly illustrated by Clark (1983), neither are the relations random. I will argue that higher structural relations within and across concepts, as well as patterns of conceptual change, can have major and systematic influences on word meanings themselves. In this chapter I explore four aspects of concepts and conceptual change that may be particularly relevant to our understanding of semantic development: (a) interdomain differences, (b) causal beliefs versus atheoretical tabulations, (c) differentiation versus sharpening, and (d) the effects of a changing conceptual base on lexical induction.

Starting at about 18 months of age, children become remarkably capable of learning the vocabulary of natural languages. Yet word learning presents a problem of induction that must somehow be solved by such very young children, with their limited information processing abilities. In order for children to acquire language as rapidly as they do, they must have biases that enable them to rule out many alternative hypotheses for the meanings of a word and that lead them instead to focus on hypotheses that are reasonably likely to be correct (Markman, 1989; Markman & Hutchinson, 1984; Markman & Wachtel, 1988). A sophisticated, intelligent adult, let alone a 2-year-old, would never be able to settle on the meaning of a word by open-mindedly considering every possible hypothesis and waiting for evidence that would rule out all but one (Quine, 1960). It has been suggested that children use several assumptions to solve the inductive problem posed by word learning; among these are the whole-object assumption, the taxonomic assumption, and the assumption of mutual exclusivity. In this chapter, I will review some of the evidence that children rely on these assumptions to guide their initial hypotheses about what words mean and will try to reconcile this argument with some conflicting evidence about whether these constraints are available to young children just starting to acquire language.

The taxonomic and whole-object assumptions

When an adult points to an object and labels it, the novel term could refer to an object category, but it could also refer to a part of the object or to its substance, color, or weight, and so on.

Languages depend for their content on words, and learning words makes up a large proportion of what children do as they acquire language. By one calculation, children learn on average nine new words a day between age 1 and age 6. And beyond age 6, by other estimates, they continue to add some six words a day to their vocabulary (see Carey, 1978; Liberman, 1989; Templin, 1957). But learning a word – its form and meaning – is no small task. It requires that one be able to identify the form of the word, its beginning and end, so that it can be picked out from the stream of speech and produced, eventually, in a form recognizable to others. And it requires that one learn what it means. This includes learning what parts of words mean, since knowing this offers children a way of expanding their current vocabulary much as adults do. When the occasion demands, children can construct new words out of known parts that they combine to convey novel meanings. They can also analyze known parts as they try to interpret new words.

But an unfamiliar word heard on a particular occasion could pick out an event or an object, a state or a relation, or a part of any of these. How does one narrow down the possibilities on each occasion in order to arrive at the conventional meaning? Children are aided in this task, I propose, by their reliance on various principles that help guide their acquisition of words. Two of these principles are general pragmatic principles that guide language use for adults as well as children.